Informacionnye Tehnologii

Eigenvectors of tridiagonal Hermitian matrices are used to model and synthesize ensembles of multi-valued orthogonal code sequences of various classes. Conditions are determined under which, within the set of ensembles of multi-valued orthogonal code sequences modeled by the eigenvectors of Hermitian matrices, distinct classes of binary, multi-level, and multi-phase code sequences are distinguished.

This article examines the concept of a human digital twin, a virtual model simulating the physical, biological, and cognitive characteristics of a real person. It focuses on analyzing the sources and types of data required to create the twin, as well as discussing its architecture and defining task classes for it. А human digital twin occupies a central place in the overall system of digital technologies, integrating the capabilities of Big Data, IoT, artificial intelligence, blockchain, augmented and virtual reality, cloud and edge computing. This allows the twin to become a key link and control center in the digital ecosystem. The article also examines the ethical issues associated with creating a human digital twin, including privacy and data security. It is noted that the twin can be used to predict human behavior and manage their state, which opens up new opportunities but also carries certain risks. The differences between a human digital twin and a machine digital twin are highlighted, stating that the human twin is a more complex system, requiring an interdisciplinary approach and consideration of multiple factors, such as genetics, biography, and psychophysiological characteristics. А multi-level architecture for a human digital twin is proposed, enabling the efficient design of data collection and integration processes, data processing, and the interaction of services and applications to support decision-making. The classes of problems that can be solved using a human digital twin and approaches to addressing them are also discussed.

The problem of modeling event flows at the input of information systems using a message broker is considered. Approaches are proposed that allow reproducing key statistical characteristics of real event flows, such as intensity, temporal structure, and seasonal fluctuations, based on both historical and real-time data. Two modeling algorithms are described: an adaptive algorithm for modeling the flow of events with a limited aftereffect, based on the Lainiotis separation method, and an algorithm based on the Markov model for changing discrete states that determine the flow parameters. Experiments have been conducted to compare the flows of events operating in a real system with a message broker and flows modeled using the proposed algorithms.

With the rapid growth in the use of credit cards in electronic payments, financial institutions and financial service providers are becoming vulnerable to fraud, which leads to huge losses every year. The development and implementation of an effective credit card fraud detection system is essential to reduce such losses. The presented paper analyzes current scientific work in the field of developing methods and models of artificial intelligence to ensure the security of financial transactions. The purpose of this paper is to review and compare machine learning models and methods for conducting secure financial transactions using credit cards. The above publications in this area mainly use a data set on fraudulent credit card transactions collected from European cardholders. It also mentions publications that use both synthetic and other datasets. Among the machine learning algorithms used in these publications, the effectiveness of decision trees, random forests, SVM, logistic regression and other methods on anonymized credit card fraud data, as well as algorithms using neural networks, is investigated and tested. The researchers apply these methods to preprocessed data samples. To assess the quality of the machine learning model, various special metrics are considered in classification tasks, such as accuracy, completeness, F-measure, etc. А comparative analysis of these publications has revealed several of the most preferred and effective methods for processing financial transactions.

The study examines the effectiveness of neural network-based methods for the synthesis of sonar images. An experimental evaluation was conducted to assess the performance of the neural style transfer method, the deep painterly harmonization method, and the arbitrary style transfer method with adaptive normalization. An original modification of the neural style transfer method is proposed, which demonstrated superior effectiveness compared to other methods in generating sonar images.